Replicate endpoint issue

[natolambert]

The api returns the following internal cuda error on the version you pinned:

  File "/Users/nathanl/miniconda3/envs/smolpod/lib/python3.10/site-packages/replicate/client.py", line 145, in run
    raise ModelError(prediction.error)
replicate.exceptions.ModelError: ('Error Running inference with local model', RuntimeError('cuFFT error: CUFFT_INTERNAL_ERROR'))

I bumped to latest:

    output = replicate.run(
        # "thomasmol/whisper-diarization:7e5dafea13d80265ea436e51a310ae5103b9f16e2039f54de4eede3060a61617",
        "thomasmol/whisper-diarization:cbd15da9f839c5f932742f86ce7def3a03c22e2b4171d42823e83e314547003f",
        input={
            "file_url": file_url,
            "num_speakers": speakers_count,
            "prompt": "Audio of Latent Space, a technical podcast about artificial intelligence and machine learning hosted by Swyx and Alessio."
        }
    )